【pytorch】在 Jupyter Notebook 中使用损失图绘制库 livelossplot

livelossplot 是一个方便的损失和精确度指标绘制工具，可以方便地在Jupyter 笔记本中实时更新当前的机器学习评价指标并绘制成曲线图。

安装方法

pip install livelossplot

基本使用方法

from livelossplot import PlotLosses
import numpy as np

liveloss = PlotLosses()

for i in range(10):
    logs = {}
    
    loss = np.random.random()
    acc = np.random.random()
    val_loss = np.random.random()
    val_acc = np.random.random()
    
    logs['log loss'] = loss
    logs['accuracy'] = acc
    
    logs['val_log loss'] = val_loss
    logs['val_accuracy'] = val_acc
    
    liveloss.update(logs)
    liveloss.send()
    

accuracy
	training         	 (min:    0.073, max:    0.883, cur:    0.802)
	validation       	 (min:    0.089, max:    0.934, cur:    0.478)
log loss
	training         	 (min:    0.023, max:    0.950, cur:    0.361)
	validation       	 (min:    0.051, max:    0.870, cur:    0.460)

也可更换任意评价指标名称，注意名字应该匹配

liveloss = PlotLosses()

for i in range(10):
    logs = {}
    
    loss = np.random.random()
    acc = np.random.random()
    val_loss = np.random.random()
    val_acc = np.random.random()
    
    logs['My_Metric'] = loss # 也可以在这更换评价指标名称
    logs['accuracy'] = acc
    
    logs['val_My_Metric'] = val_loss # 也可以在这更换评价指标名称 注意名称应该匹配
    logs['val_accuracy'] = val_acc
    
    liveloss.update(logs)
    liveloss.send()
    

My_Metric
	training         	 (min:    0.086, max:    0.893, cur:    0.867)
	validation       	 (min:    0.011, max:    0.948, cur:    0.674)
accuracy
	training         	 (min:    0.048, max:    0.939, cur:    0.048)
	validation       	 (min:    0.081, max:    0.934, cur:    0.366)